This course teaches rigorous mathematical and algorithmic techniques to enable safety design and analysis in autonomy, including the tasks of controller training, system modeling, requirements specification, and safety verification. The class offers an opportunity to dive into a broad range of autonomy applications, from unmanned drones to medical devices.
The course material consists of four modules:
The course material consists of four modules:
- Autonomy modeling: representing uncertain autonomous systems with Markov decision processes.
- Temporal safety specification: temporal logics to specify safety properties.
- Reinforcement learning for control: applied algorithms to train autonomous controllers in simulation.
- Probabilistic verification: algorithms/tools to compute safety guarantees for probabilistic models.
This course is suitable for students who want to gain state-of-the-art knowledge in safe autonomy, deepen their understanding of rigorous and learning-enabled design and validation of complex systems, gain practical experience with a particular type of autonomous systems, or complement their research with safety assurance.
Credits: 3
Teaching assistant: Yuang Geng
See the syllabus for details.
Update: check out the highlights of student projects!
If you are interested in the teaching materials, please reach out.